x
How Should LLM Agents Spend Their Test-Time Compute — LessWrong